Whatnot × Anthropic Interactive Value Model

A worked example, built so Whatnot's team can drive it themselves -- not a pitch deck.

Does the business case clear the cost of the technology?

Capability doesn't monetize -- adoption does. Move the operational metrics and the model converts each into a dollar, nets it against an enterprise agreement, and tells you the moment Whatnot is in the green.

How to read every input

Sourcedpublic & linked Discoveryyour number to calibrate in week one Illustrativeenterprise pricing is negotiated, not public

The business case

Year-1 realized business case

Adoption-adjusted value captured in the first 12 months, minus first-year total cost.

Year-1 realized netYear-1 realized value - Year-1 total cost

Pays for itself inFirst month where cumulative value >= cumulative cost

Full-adoption annual run-rateFull annual value - recurring annual platform cost

Return on recurring costFull annual value / recurring annual platform cost

0×

Year 1 uses the adoption-ramp assumption. Payback shows when cumulative realized value catches cumulative cost. Full adoption shows the annualized run-rate after the model reaches steady state.

The software does not automatically pay for itself. Adoption pays for it.

Verdict most sensitive to

Year-1 realized valueBreak-even line = first-year total cost

Cost

Green: Year-1 case clears Adoption-adjusted value exceeds first-year total cost. Yellow: Long-term case clears, Year-1 still ramping Steady-state run-rate is positive, but Year-1 adoption does not yet cover first-year total cost. Red: Case does not clear Current adoption and cost assumptions do not support the investment.

Where the value comes from: revenue leads, cost-to-serve funds

Retained GMV revenue Land

sellers at risk × GMV/seller × retention uplift × take rate × margin

Incremental intl. revenue Expand

reachable GMV × translation lift × take rate × margin

Cost-to-serve avoided Support

annual support + moderation cost × automation reduction

Drive the model: every input is a question for discovery, not a claim

Land

Seller support & SPS

Support rating → Premier status → retained GMV. SPS bar (95% / 2 days) is the published gate.

Premier sellers at risk Disc.

GMV / Premier seller / yr Disc.

Retention uplift Disc.

Expand

Live-stream translation

Real-time translation opens non-English buyers, creating net-new international GMV.

Reachable GMV ($M) Disc.

Conversion lift Disc.

Benchmark: AI chat lifted beauty conv. ~3–4% to ~12% (src)

Lever

Take rate & cost

Commission converts GMV → revenue. Automation cuts support & moderation cost.

Take rate Src

Whatnot base 8% commission (src)

Annual support cost ($M) Disc.

Automation reduction Disc.

Cost

Enterprise cost stack

Enterprise cost is not just software licensing. A credible first-year case should separate recurring platform cost from upfront implementation cost.

Recurring annual platform cost

Claude for Enterprise seats Ill.

Internal users across engineering, operations, trust & safety, support, marketing, analytics, and enablement.

Committed API / platform usage Ill.

Product-embedded realtime translation, vision, moderation, listing generation, seller copilot, buyer discovery, and support automation.

Ongoing monitoring / evaluation Ill.

Model testing, quality assurance, hallucination checks, escalation paths, policy governance, and performance monitoring.

Ongoing maintenance / enablement Ill.

Internal playbooks, workflow updates, adoption management, training refreshes, and optimization.

Recurring annual platform cost

Upfront implementation cost

Data ingestion / RAG setup Ill.

Company policies, seller support content, product taxonomy, trust & safety documentation, support macros, and historical case knowledge.

Internal integrations Ill.

Connection to support tooling, analytics, marketplace data, policy systems, and internal documentation.

Security / legal / procurement Ill.

DPA, SSO, permissions, audit review, legal review, vendor approval, and internal risk assessment.

Internal engineering allocation Ill.

Product implementation, QA, evals, prompt systems, workflow embedding, and technical deployment.

Change management and training Ill.

Department playbooks, stakeholder onboarding, adoption tracking, and team training.

Upfront implementation cost

First-year total cost: $0

Adoption and cost reality checks

Adoption is the bridge between cost, payback, and run-rate.

The software does not automatically pay for itself. Adoption pays for it. Year-1 economics use a monthly adoption ramp; full-adoption economics show steady-state run-rate once the platform is embedded into operating workflows.

How much of new revenue is real profit Disc.

GMV revenue is not profit. This margin applies only to retained and incremental GMV revenue. Cost-to-serve savings are already counted as net value.

Starting adoption / value capture Disc.

Initial value capture during early rollout.

Target adoption by Month 12 Disc.

Modeled adoption level by the end of the first year.

Full-adoption steady-state target Disc.

Annualized run-rate once the platform is embedded into operating workflows.

Deploy

Upfront implementation cost begins.

Adopt

Partial value capture compounds through workflows.

Break even

Cumulative value equals cumulative cost.

Scale

Full-adoption annual run-rate becomes the steady-state view.

Adoption sensitivity

Same value and cost formulas, different adoption realized. This shows why the commercial question is adoption velocity, not model capability alone.

Adoption realized	Value captured	Year-1 net	Payback timing	Status

The math, shown -- not a black box

Retained GMV revenue

Incremental international revenue

Cost-to-serve avoided

Net business case

Where adoption shows up

Who actually touches the technology, and how the two cost lines earn out.

The seat line funds internal teams; the API line funds product features your sellers and buyers use. Adoption here is what turns committed spend into renewal.

Live at Whatnot today (sourced) Proposed application (illustrative)

Seat line

Internal teams: Claude for Enterprise

Seat-based, no-training-on-data, SSO, connectors, and custom Claude projects grounded in Whatnot's own docs and style.

Trust & Safety: fraud and abuse triage, policy drafting, case summaries. Already augmenting the rule engine with LLMs + OCR.
Seller Support & Ops: agent copilot, macro generation, SPS-readiness coaching for sellers.
Engineering: Claude Code for code, internal tooling, and faster shipping.
Marketing & Content: campaign copy and localization from brand ground-truth.
Data & Analytics: natural-language querying and report drafting.
Knowledge / Enablement: RAG over policies, docs and style so any employee references one source of truth.

API line

End users: product-embedded API

Realtime, vision, and moderation models embedded in the live-shopping product itself.

Sellers

Live-stream copilot: real-time buyer Q&A and pricing prompts mid-show.
Real-time translation: opens non-English buyers to every stream.
Listing generation: titles, descriptions and tags from a photo (vision).
Auto-moderation: OCR image-fraud detection, human-in-the-loop. Live today.

Buyers

Conversational discovery: natural-language search across live and catalog.
Personalized recommendations: surface the next show or drop to buy.
Self-serve support: order, dispute and return resolution in-app.

Methodology & sources

What's sourced, what's modeled, what's yours to calibrate.

This model separates three views: first-year realized economics, payback timing, and full-adoption annual run-rate. Year-1 value is adjusted by the adoption ramp because teams and product workflows do not reach full usage on day one. Payback is the month when cumulative realized value exceeds cumulative total cost. Full adoption represents the annualized run-rate after the platform is embedded into operating workflows.

Sourced (linked)

Public sources are used for directional benchmarks: Whatnot's published 8% base commission, the SPS bar, reported GMV/category context, and public engineering discussion of existing trust & safety AI use. Source links remain below.

Discovery (placeholders)

Sellers at risk, GMV/seller, retention uplift, reachable international GMV, support cost, automation reduction, contribution margin, and adoption ramp are placeholders. They should be replaced with Whatnot's actual operating data during discovery.

Illustrative enterprise cost

Actual Anthropic enterprise pricing, API commitments, implementation scope, and renewal terms would be negotiated. The cost model is illustrative and should be replaced with Whatnot's actual commercial terms during discovery.

Economics logic

Year-1 realized net uses adoption-adjusted value captured in the first 12 months minus first-year total cost. Full-adoption annual run-rate uses steady-state annualized value minus recurring annual platform cost. Payback month is calculated from cumulative realized value versus cumulative total cost.

Sources

• Whatnot Seller Fees & Commissions Schedule: 8% base commission
• Whatnot Seller Provided Support: 95% / 2-business-day bar, Sept 1 2026
• Whatnot 2026 State of Live Selling Report: $8B GMV, category growth
• Whatnot Engineering: existing LLM trust & safety + OCR fraud
• AI-commerce conversion benchmarks: illustrative lift range only